Presentation of the Goodpasture autoantigen to CD4 T cells is influenced more by processing constraints than by HLA class II peptide binding preferences.

Class II molecules are believed to influence immune responses by selectively binding antigen-derived peptides for recognition by T cells. In Goodpasture's (anti-glomerular basement membrane) disease, autoimmunity to the NC1 domain of the alpha3-chain of type IV collagen (alpha3(IV)NC1) is strongly associated with HLA-DR15. We have examined the influence of the peptide binding preferences of DR15 molecules on the selection of alpha3(IV)NC1-derived peptides displayed bound to DR15 molecules on the surface of alpha3(IV)NC1-pulsed DR15-homozygous Epstein-Barr virus-transformed human B cells. The preferences of DR15 molecules were investigated using a panel of 24 overlapping peptides spanning the sequence of alpha3(IV)NC1. The alpha3(IV)NC1-derived peptides selected for display to T cells were determined by biochemical analysis as reported previously (Phelps, R. G., Turner, A. N., and Rees, A. J. (1996) J. Biol. Chem. 271, 18549-18553). Three nested sets of naturally presented alpha3(IV)NC1 peptides were detectable bound to DR15 molecules. Peptides representative of each nested set bound to DR15 molecules, but almost two-thirds of the alpha3(IV)NC1 peptides studied had as good or better DR15 affinity than those identified as naturally processed. Thus alpha3(IV)NC1 presentation to T cells is determined more by "processing factors" than by the preferences of relatively indiscriminate DR15 molecules. The results have important implications for the use of class II peptide binding data to aid identification of potential T cell epitopes, especially for antigens which, like alpha3(IV)NC1, contain many sequences able to bind class II molecules.

Antigen presenting cells (APC) 1 potentially exert a profound influence on immune responses, including those to self antigens, because they regulate the way T cells recognize antigens. CD4 T cells recognize antigens in the form of processed peptides presented bound to MHC class II molecules on the surface of class II positive APC types (1,2). These include cells important in the initiation of immune responses, such as dendritic cells, B cells, and macrophages, and cells important in regulating the T cell repertoire and self-tolerance, such as thymic epithelia. How APC select antigen-derived peptides for display to T cells therefore not only constrains the peptide specificity of responding T cells, but also influences the repertoire of T cells available to mount immune responses. Furthermore, the way APC present antigens may determine the immunodominant T cell response, which at least for some exogenous antigens is directed at the antigen-derived peptide displayed at the highest level (3). There is therefore great interest in understanding how APC generate antigen-derived peptides and make a selection for display to T cells, both as an approach to identifying T cell epitopes for specific immune modulation and toward understanding the basic biology of immune responses.
APC, like many cell types, internalize extracellular proteins into their endosomal/lysosomal pathway where denaturing (low pH and reducing) conditions promote the access of endosomal proteases that degrade or "process" protein antigens into peptides and, ultimately, to amino acids and di-peptides (4,5). APC that express MHC class II genes target newly synthesized MHC class II molecules to processing compartments where they become competent to bind antigen-derived peptides and intact denatured proteins (6). Peptides able to form stable complexes with class II molecules are protected from further proteolysis (7,8) and may be transported to the cell surface for recognition by CD4 T cells. Accordingly, the selection of antigen-derived peptides displayed by APC is thought to be determined by an interplay between antigen denaturation and proteolysis (processing) on the one hand and MHC class II/peptide binding on the other. Antigen processing is poorly understood, and the range of antigen-derived peptides generated and made available for binding class II molecules is not yet known for any antigen. Few of the probably numerous endosomal proteases have been identified and their specificities characterized. Furthermore, how an antigen is processed is likely to depend not only upon its amino acid sequence but also on its tertiary and quaternary structure, which may influence susceptibility to denaturation and the access of endosomal proteases (9 -11). In contrast, the way MHC class II molecules bind to peptides is well understood, and the ligands preferred by a particular class II molecule can fairly reliably be predicted.
Class II molecules are heterodimeric membrane glycoproteins with an extracellular peptide binding groove (12). Peptides are clasped with nine amino acid residues (the core binding sequence) accommodated in the groove, lying in extended conformation. Binding is largely stabilized by sequence nonspecific interactions with main chain nitrogen and oxygen atoms in the bound peptide. However, stable binding requires peptide side chain residues that point into the floor and walls of the groove to be accommodated in complementary pockets, the size and chemical character of which are extensively influenced by polymorphisms in class II alleles. Consequently, class II molecules bind a wide range of peptides but have allele-specific preferences for peptides with core binding sequences containing appropriately spaced, efficiently accommodated side chain residues (13). Peptide binding preferences can be described using peptide binding motifs (14), and motifs have been developed for some class II molecules which fairly reliably predict the class II affinity of a peptide from its amino acid sequence (15)(16)(17). Thus while it is currently impossible to predict how an antigen will be processed by APC, it is relatively straightforward to identify the antigen-contained peptides which, if generated by processing, should be preferentially bound by class II molecules and hence selected for presentation to T cells. Antigen-contained sequences with high class II affinity can be identified by comparing the class II binding affinity of synthetic peptides with overlapping sequences spanning the sequence of an antigen or by using peptide binding motifs. Both approaches have been successful in identifying immunologically relevant T cell epitopes (16 -19).
Goodpasture's disease is an autoimmune nephritis caused by autoimmunity to the 235 amino acid C-terminal NC1 domain of type IV collagen (␣3(IV)NC1) (20,21), a component of basement membrane in some tissues. Almost 80% of patients carry a haplotype bearing HLA DRB1*1501 and DRB5*0101 which encode the ␤-chains of the HLA class II molecules DR15b 2 and DR15a, respectively (22). We are examining how DR15-homozygous human APC process and present ␣3(IV)NC1 as an approach to identifying potential T cell epitopes and toward understanding how antigen processing and presentation and HLA genes influence susceptibility to autoimmune diseases. Our approach has been to biochemically characterize DR15associated peptides, purified from DR15-homozygous EBVtransformed human B cells, using matrix-assisted laser-desorption time of flight mass spectrometry and reverse phase HPLC (23). Comparison of peptides purified from ␣3(IV)NC1pulsed and control non-pulsed APC identified 17 extra putative ␣3(IV)NC1-derived peptides. The difficulty with this approach is the low frequency of antigen-derived peptides; all the extra peptides occurred at low level within complex peptide mixtures, and their sequences could not be directly determined. However, by exploiting the tendency of class II-associated peptides to occur as nested sets, we were able to define the sequences of eight extra peptides. They comprised two nested sets centered on the common core sequences LFCNVNDVCNF and LEE-FRASPF (24).
Here we have used overlapping synthetic peptides spanning the sequence of ␣3(IV)NC1 and matrix-based peptide binding motifs to identify all sequences within ␣3(IV)NC1 able to bind to DR15 molecules, and we ranked them according to DR15 affinity. Surprisingly, the peptides previously shown to be naturally processed from ␣3(IV)NC1 and selected for presentation bound to DR15 molecules have only intermediate DR15 affinity. Because DR15 molecules were expected to preferentially bind the available ␣3(IV)NC1 sequences for which they have greatest affinity, the biochemically detectable DR15-associated putative ␣3(IV)NC1-derived peptides purified from ␣3(IV)NC1pulsed APC were re-analyzed for the possible occurrence of ␣3(IV)NC1 sequences with high DR15 affinity. The analysis identified a third nested set of naturally presented ␣3(IV)NC1derived peptides. However, the most striking result was that most ␣3(IV)NC1 peptides with high DR15 affinity were unde-tectable among DR15-associated peptides purified from ␣3(IV)NC1-pulsed APC. Indeed, most of the naturally presented ␣3(IV)NC1-derived peptides bound to DR15 molecules have only intermediate affinity for DR15 molecules, both in the range of affinities measured for ␣3(IV)NC1-derived peptides and in the range of affinities reported for DR-associated peptides. Thus the peptide binding characteristics of DR15 molecules appear to be largely permissive in determining which ␣3(IV)NC1-derived peptides are displayed by DR15 APC, presumably sub-dominant to unknown processing factors.

EXPERIMENTAL PROCEDURES
Synthetic Peptides-A panel of synthetic peptides was made with sequences spanning the sequence of ␣3(IV)NC1 (Table I). They were purified by HPLC and had their composition confirmed by matrixassisted laser-desorption time of flight mass spectrometry. All the peptides were prepared as ϳ1 mM solutions in 20% Me 2 SO, 5 mM dithiothreitol and stored at 4°C because some had limited solubility in water and some tended to form disulfide-linked dimers. Stock peptide solutions were subjected to amino acid analysis to measure peptide concentrations more accurately. The peptides hemagglutinin 307-319 and myelin basic protein 86 -98(98A), called HAP and MBPP respectively, were biotinylated at the N-terminal to make the labeled reference peptides designated *HAP and *MBPP.
Peptide retention times were measured using HPLC conditions identical to those described previously (23). Predicted retention times were calculated essentially as before, but incorporating retention time data for 11 more synthetic peptides, including peptides that overlapped all the ␣3(IV)NC1 sequences for which predicted retention times were calculated. Measured and predicted retention times exhibited a good fit to a linear model (R 2 ϭ 0.93). Nonlinearity, for example due to secondary structure formation, was probably minimal for the sequences for which prediction was attempted as they all overlapped by at least 10 residues a synthetic peptide whose measured retention time was close to its predicted retention time.
Purification of Class II Molecules-HLA-DR15a and -DR15b molecules were affinity purified from the murine L cell transfectants LDR2a and LDR2b, respectively (25), because available monoclonal antibodies do not distinguish between the two DR molecules carried by human DR15 homozygous APC. DR15a is composed of the gene products of DRA and DRB5*0101 and DR15b of the gene products of DRA and DRB1*1501. Cells were lysed in 1% Nonidet P-40, 150 mM NaCl, 50 mM Tris, pH 8, containing protease inhibitors and DR molecules affinity purified employing the monoclonal antibody L243 conjugated to cyanogen bromide-activated Sepharose.
Peptide Binding Assays-Peptide binding to purified class II molecules was measured using an inhibition assay based on that described by Tompkins et al. (26). The myelin basic protein peptide MBP86-98(98A) and flu hemagglutinin peptide HA307-319 were selected as reference peptides because of their reported high affinity for DR15b and DR15a, respectively (27). The test peptides were incubated at a range of concentrations with affinity purified DR molecules (Ϸ0.5 M) and labeled reference peptides in 0.5% Nonidet P-40, 150 mM NaCl, 0.1 M sodium citrate, pH 5.5, in the presence of protease inhibitors for 16 h at 26°C. The reference peptides were used at concentrations (160 nM for *MBPP and 86 nM for *HAP) at which much less than 10% bound to DR molecules. DR-reference peptide complexes were assayed by immobilizing DR-peptide complexes onto enzyme-linked immunosorbent assay plates coated with L243, labeling immobilized biotin with a streptavidin-Europium conjugate (Wallac Oy, Finland), and measuring Europium by time-resolved fluorescence at 613 nm (excitation at 340 nm). The concentrations of peptides [P i ] causing 50% reduction in reference peptide binding (IC 50(i) ) were extracted from experimental binding data by curve-fitting to functions of the form B/(1 ϩ [P i ]/IC 50(i) ) where B is binding in the absence of unlabeled test peptide (the equation derives from the law of mass action). IC 50 values were determined in at least three independent experiments. Expressed as logarithms, the experimental data were normally distributed with S.D. 0.165, and the 95% confidence limits for the arithmetic means (three measurements) were Ϯ0. 19. Therefore the 95% confidence limits for measured IC 50 values (in moles) were between (mean/1.5) and (mean ϫ 1.5).
To compare measurements of affinity with those of other researchers, it was necessary to determine an absolute measurement of peptide class II affinity (IC 50 measurements are dependent upon the experimental conditions used). When both reference peptide (P ref ) and test peptides (P i ) are used in excess of available class II, the (inhibition) dissociation 2 The terms DR15a and DR15b refer to class II molecules composed of the DR ␣-chain and DR ␤-chains encoded by DRB5*0101 and DRB1*1501, respectively. This terminology echoes the more widely used terms DR2a and DR2b and is preferred here to refer specifically to DR molecules encoded by particular DRB5 and DRB1 alleles. This was necessary because inheritance of any of several DRB1 alleles (B1*15xx and B1*16xx) gives rise to display of DR molecules indistinguishable by the serological techniques used to define the DR2 specificity.
values of *HAP/ DR15a and *MBPP/DR15b were estimated to be 5 and 16 nM, respectively, by Hill transformation of binding measured at a range of concentrations of reference peptides. Therefore, for comparison purposes, K i ϳIC 50 /11 for DR15b inhibition assays and IC 50 /18 for DR15a assays.
Other procedures were as described previously (23).

RESULTS
To examine the importance of the peptide binding characteristics of DR15 molecules in shaping the selection of ␣3(IV)NC1 peptides displayed to T cells by DR15 homozygous APC, we sought to identify all peptides within the sequence of ␣3(IV)NC1 with good affinity for DR15 molecules and analyze the results in the light of our previous biochemical characterization of naturally processed ␣3(IV)NC1-derived peptides (23). Because that study examined peptides eluted from DR15 molecules purified from ␣3(IV)NC1-pulsed MGAR cells (a line of EBV-transformed human B cells homozygous for DR15), and hence peptides that could have been bound to DR15a or DR15b molecules, it was necessary to examine peptide binding to both DR15a and DR15b molecules separately. Two complementary approaches were used. First, a set of synthetic peptides was made with overlapping sequences spanning the sequence of ␣3(IV)NC1. They were compared for their capacity to inhibit the binding of labeled reference peptides to affinity purified DR15a and DR15b molecules. Second, the sequence of ␣3(IV)NC1 was examined for peptides with high predicted affinity for DR15a or DR15b molecules using matrix based motifs to identify any peptides likely to bind well but inadequately represented within the set of overlapping peptides.
Most ␣3(IV)NC1 Peptides Bind Detectably to DR15a/b-Synthetic peptides were made with the sequences shown in Table  I. Peptides P1 to P23 are 20 to 21 amino acids in length (except P23) and span the sequence of ␣3(IV)NC1 with a mostly 10residue overlap. This degree of overlap was chosen to slightly exceed the length of bound peptide accommodated within the class II peptide binding groove (12). Peptide P3 proved difficult to synthesize in any quantity and poorly soluble in aqueous buffers, so P3 was substituted by peptides P3a and P3b.
Seventeen of the 24 peptides examined bound to DR15b molecules with varying degrees of affinity ( Fig. 1). Four of the peptides bound with high affinity and 13 with intermediate affinity, defined by the previously reported range of affinities for peptide-class II interactions (see legend to Fig. 1). The assay conditions permitted measurement of IC 50 for peptides with IC 50 Ͻϳ100 M, so did not distinguish low affinity binding (IC 50 Ͼ100 M and K i Ͼ6 -12 M) from non-binding.
Seventeen peptides also bound to DR15a molecules, but a higher proportion (9/24) bound with high affinity. This could be because the ␤-chain of DR15a has a glycine at position ␤86 which is smaller than the valine at ␤86 in DR15b and permits the P1 pocket of DR15a to more easily accommodate bulky hydrophobic side chains, especially aromatic residues, further stabilizing peptide binding (28). All but one of the high DR15a affinity peptides (P10) contain aromatic amino acids at posi-  tions able to engage in the P1 pocket of the DR15a peptide binding groove, whereas P10 has suitably placed methionine residues.
Class II Binding Motifs Identify Probable Core Binding Sequences Used by High Affinity Peptides-The sequence of ␣3(IV)NC1 was also examined using matrix-based DR15 peptide binding motifs. This was done first to identify the probable core binding sequences of the ␣3(IV)NC1 peptides with high affinity for DR15 molecules, and second to ensure all ␣3(IV)NC1 sequences with high DR15 affinity had been identified. While every nine amino acid core binding sequence in ␣3(IV)NC1 was represented within the set of synthetic peptides, it was possible that different ␣3(IV)NC1 peptides would bind to DR15 with higher affinity, perhaps through making fewer unfavorable interactions outside the groove.
Predicted DR15a binding affinity was calculated for each of the 226 nine amino acid core sequences in ␣3(IV)NC1 using the methods and matrix coefficients of Marshall et al. (17) (Fig. 2a). Thirteen core sequences were identified with predicted DR15a binding affinity better than 10 nM. Twelve of the 13 occurred within the sequences of the four synthetic peptides (peptides P4, P9, P15, and P20) with highest measured DR15a affinity, indicating a good correspondence between the prediction algorithm and experimental results for ␣3(IV)NC1 peptides. The measured and predicted DR15a affinities of the other ␣3(IV)NC1 peptides also showed a close correspondence (data not shown). The other core sequence with predicted DR15a binding affinity better than 10 nM was LQRFTTMPF which was within peptide P6 which bound DR15a with only intermediate affinity.
Similar matrix-based motifs for DR15b are not yet published preventing an exactly similar analysis of ␣3(IV)NC1 binding to DR15b. To identify the sequences in ␣3(IV)NC1 with highest predicted affinity for DR15b, we used an unpublished matrixbased motif in collaboration with T. Sturniolo and co-workers. 3 The motif utilized binding data acquired as previously reported (16), but predicted binding was calculated using a refined algorithm as recently described (15). The three core sequences in ␣3(IV)NC1 with highest predicted DR15b affinity are shown in Fig. 2b. One of these, FIMFTSAGS, occurred within the synthetic peptide with highest measured DR15b affinity (P15). The other two core sequences were well represented within peptides (P3b and P6) that had intermediate affinity for DR15b molecules (ranked 6th and 8th out of 24, respectively).
Importantly, neither motif analysis identified ␣3(IV)NC1 peptides with high predicted DR15a/b affinity that were not well represented in the set of overlapping peptides, so the measured DR15 affinities of the set of synthetic peptides could reasonably be used to rank ␣3(IV)NC1 peptides according to DR15a or DR15b affinity. The four peptides with highest affinity for DR15a and for DR15b are shown in Table II. The P1 residues of their probable core binding sequences identified by the motif analysis are underlined.
DR15 Affinity of Major Naturally Processed and Presented ␣3(IV)NC1-derived Peptides-Next we examined the DR15 affinity of naturally processed and presented ␣3(IV)NC1 peptides, to assess the degree to which their selection was dictated by the peptide binding preferences of DR15 molecules. Our previous biochemical analysis of DR15-associated peptides pu-  17. The measured affinities of the four synthetic peptides overlapping the core sequences with highest predicted affinity are superimposed as bars depicting the portion of the sequence of ␣3(IV)NC1 included and their measured affinity IC 50 /18. All the core sequences with highest predicted affinity (Ͻ10 nM or Ͼ8 on the scale shown) occur in the four peptides in the set P1-23 with highest measured affinity for DR15a, except for one which overlaps peptide P6. b, HLA DR15b: the first column shows the three core sequences within ␣3(IV)NC1 with highest predicted affinity for DR15b, identified using a matrix motif (15). Short ␣3(IV)NC1-contained peptides (2nd column) containing these core sequences (but with cysteine substituted with alanine) bound well to DR15b (3rd column). The peptides in the set P1-23 best representing these core sequences are shown in the 5th column. To permit comparison of binding data for these peptides with those for the peptides in the 2nd column (values in the 3rd column were measured in a different laboratory), binding affinities are shown relative to peptide P15 in column 6 and relative to the closely related peptide FSFIMFTSAGSEG in column 4. The peptide FSFIMFTSAGSEG (and peptide P15) had high affinity for DR15b, but the two peptides with next highest predicted affinity bound only about one-fifth as well. Superscript 1, relative to affinity measured for peptide FSFIMFTSAGSEG; superscript 2, relative to affinity measured for the peptide P15; superscript 3, a natural cysteine substituted with alanine. rified from ␣3(IV)NC1-pulsed DR15 homozygous APC identified two nested sets of naturally processed and presented ␣3(IV)NC1-derived peptides centered on the core sequences LFCNVNDVCNF and LEEFRASPF (23). The techniques used to identify these peptides had limited sensitivity so they are likely to be among the most abundant naturally presented ␣3(IV)NC1 peptides displayed bound to DR15 molecules.
First we looked for naturally presented sequences within the overlapping set of ␣3(IV)NC1 peptides. The core sequence LFCNVNDVCNF occurs in peptide P7 and LEEFRASPF in peptide P17. Neither synthetic peptide had high affinity for DR15a or DR15b (Fig. 1). P7 bound with intermediate affinity to both class II molecules and P17 bound with intermediate affinity to DR15b and undetectably to DR15a. However, neither P7 nor P17 had sequences identical to naturally presented ␣3(IV)NC1 peptides. We therefore also measured DR15a/b binding of synthetic ␣3(IV)NC1 peptides with naturally presented ␣3(IV)NC1 sequences, using the same assay (Table III). The peptide ALASPGSCLEEFRASPFLE bound to DR15b with 8-fold higher affinity than P17, still within the intermediate range of affinities. The peptide PFLFCNVNDVCNFASR bound to DR15a/b with lower affinity than P7, still within the intermediate range of affinities.
Thus quantitatively major DR15-associated naturally processed ␣3(IV)NC1-derived peptides have intermediate rather than high affinity for DR15 molecules. This was the case both in relation to measured affinities of ␣3(IV)NC1-derived peptides and by comparison to DR-affinities reported for DR-associated peptides.
Occurrence of ␣3(IV)NC1 Peptides with High Affinity for DR15 Molecules Among Naturally Processed DR15-associated Peptides-Processed ␣3(IV)NC1-derived peptides with high DR15 affinity would be expected to be more efficiently loaded onto DR15 molecules than peptides with intermediate affinity, so it was surprising that none of the previously identified naturally presented ␣3(IV)NC1 peptides contained high DR15 affinity sequences. In that experiment nine extra putative ␣3(IV)NC1-derived DR15-associated peptides were identified whose sequences could not be determined (called extra peptides because they were detected among DR15-associated peptides purified from ␣3(IV)NC1-pulsed APC but not from shampulsed APC). We therefore re-analyzed the data from that experiment to see if any of the nine unaccounted extra peptides could be high DR15 affinity ␣3(IV)NC1 peptides.
The extra peptides were characterized by mass and retention time during reverse phase HPLC. Eight of the nine unaccounted extra peptides had masses consistent with high DR15 affinity ␣3(IV)NC1 sequences overlapping those in Table II (see  Table IV). However, four were indistinguishably well matched by mass to more than one (up to five) high DR15 affinity candidate ␣3(IV)NC1 sequence, reflecting the limited precision of mass measurement (95% confidence limits Ϯ0.05%). Inspection of the candidate ␣3(IV)NC1 sequences identified three potential nested sets centered on core binding sequences with high DR15 affinity (two are shown in Table V). Such nested sets are typical of class II-associated peptides, probably because bound peptides are trimmed by exopeptidases (29), but their recognition did not distinguish between the many candidate sequences as the potential nested sets made repeated use of some of the putative ␣3(IV)NC1 peptides.
To test the sequences proposed for the extra peptides some were made as synthetic peptides, and their retention times during reverse phase HPLC were compared with those of the respective extra peptides (known from the numbers of the fractions in which they were identified). Synthetic peptides made with the three candidate sequences comprising the nested set with common core sequence GTVPLYSGFSFLFVQ (proposed P1 residue underlined) all had measured retention times within 1 min of those of the respective extra peptides (Table V). To assess the significance of matching measured retention times, we estimated the frequency with which synthetic peptides eluted by chance alone during a specific 3-min window (i.e. into a particular 1-min fraction or into either adjacent fraction) by chance alone. Presuming the retention times of peptides of random composition distribute evenly between 15 and 40 min, the interval during which most ␣3(IV)NC1 peptides we have studied elute under the chromatography conditions, then about 16% are expected to elute in any particular 3-min window, and the chance that three synthetic peptides elute during specified 3-min windows by chance alone is about 0.004. Therefore matching measured retention times strongly support (p ϳ0.004) the identification of a third nested set of naturally processed and presented ␣3(IV)NC1derived peptides centered on the core sequence GTVPLYSGFS-FLFVQ (Table V). By contrast, synthetic peptides made with four candidate sequences containing the core sequence LEPY-ISRCTV all had measured retention times substantially (11-12 min) different to those of the respective extra peptides, refuting these candidate sequences. TABLE II ␣3(IV)NC1 peptides with highest affinity for DR15a/b molecules Peptides in ␣3(IV)NC1 with the highest affinity for DR15 molecules were identified using a set of overlapping peptides and ranked according to DR15a/b binding with respect to labeled peptide, as described under "Experimental Procedures." Motif analysis suggests that the peptides bind in the class II peptide binding groove with the side chain of (one of) the underlined residues engaged in pocket P1. Class

Natural Processing of Goodpasture Antigen 2
Retention time analysis was extended to all the candidate sequences by calculating predicted retention times (Table VI). The 95% confidence limits for predicted retention times were between Ϯ3.7 min (at 12 min) and Ϯ3.5 min (at 19 -39 min). Barring extreme nonlinearity, predicted retention times refuted all but two of the remaining high DR15 affinity candidate ␣3(IV)NC1 sequences. No significance was attached to the two candidate sequences with matching predicted retention times because they did not share a common core sequence, and the estimated likelihood of predicted retention times matching a particular 8-min window (3.5 min either side of the 1-min fraction in which an extra peptide was identified) by chance alone was estimated to be about 0.3. Therefore in calculating predicted retention times for the 11 candidate sequences for which measured retention times were not available, 3 are expected to have matching predicted retention time by chance alone.

DISCUSSION
The results presented here and in Ref. 23 identify three nested sets of ␣3(IV)NC1 peptides naturally processed from intact ␣3(IV)NC1 and selected for display bound to DR15 molecules by DR15 homozygous EBV-transformed human B cells (Table VII). The peptides of the nested set centered on the core sequence LEEFRASPF must be presented bound to DR15b molecules, the DR molecule most likely to confer susceptibility to Goodpasture's disease (22), because they have undetectable affinity for DR15a. The other two nested sets of peptides bind with intermediate or high affinity to DR15a and DR15b, so could be presented bound to either (or both) DR15 molecule. The three nested sets of peptides are likely to be the most abundant ␣3(IV)NC1 peptides displayed by the APC used in the experiments because the techniques used to identify them were of limited sensitivity. This enables us to investigate the factors that determine the ␣3(IV)NC1 peptides selected for display to T cells. Here we have examined the influence of the peptide binding preferences of DR15 molecules.
The binding data show that almost 80% of the ␣3(IV)NC1 peptides examined bound to DR15a or DR15b with intermediate or high affinity. To assess the degree to which the peptide binding characteristics of DR15 molecules influenced the selection of ␣3(IV)NC1 peptides displayed by DR15-homozygous APC, the ␣3(IV)NC1 peptides were ranked according to DR15 affinity. Strikingly, the synthetic peptides representative of the naturally processed peptides ranked only between 4th and 14th within the set of 24 ␣3(IV)NC1 peptides (Table VIII). For example peptide P7, representative of the nested set centered on LFCNVNDVCNF, had the 9th highest affinity to DR15a and 5th highest affinity to DR15b. These data suggest that binding affinity for DR15 molecules has a relatively minor influence on which ␣3(IV)NC1 peptides are displayed on the surface of APC. However, with the exception of P3b, peptides in the overlapping set were not identical to naturally processed peptides. P7 and P17 included the entire common core sequences of naturally processed nested sets of peptides, but it was possible that they bound less well to DR15 molecules than naturally processed peptides because of interactions outside the peptide binding groove. This explanation was confounded by the additional experiments in which synthetic peptides with naturally processed sequences were shown to bind to DR15 molecules with only intermediate affinity. Thus the major (biochemically detectable) ␣3(IV)NC1 peptides selected for presentation bound to DR15 molecules are not those in ␣3(IV)NC1 with highest DR15 affinity, indeed naturally presented peptides bind to DR15 molecules no better than peptides representing TABLE IV Naturally processed DR15-bound peptides with masses matching ␣3(IV)NC1-derived peptides with high DR15 affinity The first column shows the fraction numbers of HPLC fractions found to contain extra putative ␣3(IV)NC1-derived DR15-associated peptides, and the second column shows their masses. Candidate sequences for the extra peptides were identified by searching the sequence of ␣3(IV)NC1 for peptides with matching calculated mass. Extra peptides for which candidate sequences were identified that contained entire high affinity core binding sequences are marked by *, placed in columns headed by the names of the peptides in the set P1-23 among which the high affinity core sequences were identified.  almost two-thirds of the sequence of ␣3(IV)NC1. This means that the peptide binding characteristics of DR15 molecules must be largely permissive to other processing factors in determining the selection of ␣3(IV)NC1-derived peptides presented on the surface of APC in our experiment. Inspecting the sequences of the proposed naturally processed peptides gives some indications of the processing factors that may have constrained the selection of ␣3(IV)NC1 peptides displayed bound to DR15. All except one of the peptides contain one or more cysteine residues likely to be involved in disulfide bonds (based on the reported structure of the highly homologous ␣1(IV) chain, see Ref. 30), and many known T cell epitopes derive from regions of proteins involved in disulfide bonds, possibly because the presence of the disulfide linkage affords some protection from proteolytic attack (31). The preparation of ␣3(IV)NC1 used in this study was reduced during purification but had ample opportunity to oxidize prior to and during pulsing of the APC. It is also striking that two of the nested sets, the sets containing LFCNVNDVCNF and LEEFRASPF, derive from corresponding portions of the two hemidomains of ␣3(IV)NC1. The content of DR15 binding motifs in these regions is not strikingly different from that of ␣3(IV)NC1 as a whole, but it is likely they have similar secondary and tertiary structure which could favor their selection for presentation to T cells. The proximity of these regions to the surface of intact ␣3(IV)NC1 is not known, but they could be similarly inaccessible to the endosomal proteases or similarly accessible for early binding (even as partially unfolded intact protein) to class II molecules (32,33) with consequent protection from proteolysis (7,8). Processing could also favor the naturally presented peptides by selectively destroying some of the ␣3(IV)NC1 peptides with higher DR15 affinity. The endosomal enzymes important in ␣3(IV)NC1 processing are unknown but presumably include those shown to process other antigens (reviewed in Ref. 5). Recently the specificity of the candidate endosomal enzyme cathepsin E has been described in detail (34) permitting a search of the sequence of ␣3(IV)NC1 for vulnerable peptide bonds. 4 Intriguingly, the two ␣3(IV)NC1 sequences most likely to be cut by cathepsin E (CPHGW ISL and KGFSF IMF) occur in peptides P14 and P15, both of which have very high DR15 affinity and neither of which is detectably presented bound to DR15 molecules.
A more mundane explanation for the lack of relationship between DR15 affinity and the selection of ␣3(IV)NC1 peptides displayed bound to DR15 is that in vitro measurements of peptide DR interaction poorly represent peptide-DR interactions within the processing compartments of APC. Certainly the conditions are different within APC; most antigen-derived peptides generated by processing bind to newly synthesized class II molecules by displacing the CLIP peptide from the peptide binding groove. The interaction is catalyzed by HLA-DM which also appears to edit the resulting class IIpeptide complexes (35,36); antigen-derived peptides become stably bound within peptide class II complexes as quickly as 30 min after addition of antigen. In our experiments high concentrations of peptide were used to drive competitive displacement of previously bound non-CLIP peptides from detergent-solubilized DR molecules in the absence of HLA-DM and over a 16-h time course. Binding assays were performed at endosomal pH because pH has been shown to influence DR-peptide interactions (37). However, there is a large body of evidence that peptide-class II interactions measured by similar techniques to those used here have immunological relevance. In particular, similar inhibition binding assays have been used to show the following: (i) that major T cell epitopes in exogenous antigens bind to the restricting class II molecules with high affinity (38,39); (ii) that the more immunogenic peptides within exogenous antigens are a subset of those with higher class II affinity (40,    VIII Rank DR15 binding of naturally processed ␣3(IV)NC1 peptides A representative peptide from each nested set of naturally processed ␣3(IV)NC1 peptides is shown next to the sequence of the most similar peptide in the set of overlapping ␣3(IV)NC1 peptides. Peptide 3b is identical to the sequence assigned to the extra peptide in fraction 37. Rank affinity was determined by ordering the peptides P1-P23 by decreasing DR15a/b affinity, omitting P17 in the case of peptides 31-2024.9 and P7 in the case of 31-1845.1. Neither P17 nor 31-2024.9 bound detectably to DR15a, so their affinities could not be ranked. 41); and (iii) that peptides eluted from purified class II molecules bind well to class II molecules in vitro (23,42,43). Therefore, even though the measurements of class II binding affinity did not model loading of class II molecules within APC, they should indicate the relative capacity of DR15 molecules to bind ␣3(IV)NC1 peptides. It is striking that ␣3(IV)NC1, the autoantigen attacked in Goodpasture's disease, contains so many sequences able to bind to DR15 molecules, the class II molecules strongly associated with this disease. It is likely that self-tolerance is best established to self-antigen-derived peptides constitutively displayed at highest level and that autoimmunity is directed at other peptides usually presented at much lower level, called cryptic epitopes (44 -46). From this standpoint, the results are intriguing in two ways. First, ␣3(IV)NC1 contains a higher proportion of peptides able to bind DR15a/b (17/24 or 71%) than the 15-25% observed for a range of exogenous antigens and human DR molecules, including DR15 (39), suggesting considerable scope for the presentation of peptides to which tolerance has been incompletely established. Interestingly, intermediate or better DR affinity was found for a high proportion of overlapping peptides spanning myelin basic protein ( 50), all known or proposed autoantigens. Second, ␣3(IV)NC1 contains several peptides (e.g. P15 and P14) with very high affinity for DR15 that are not constitutively displayed to T cells, but which under conditions of aberrant processing (e.g. extracellular processing) could be powerful immunogens. This would explain why HLA-DR15 is important in susceptibility to Goodpasture's disease even though the peptide binding characteristics of DR15 appear to have only a permissive influence on how ␣3(IV)NC1 is presented, at least under the conditions studied.
Unless ␣3(IV)NC1 is very atypical, our results suggest that identifying peptides within antigens with high class II affinity may not economically identify naturally presented peptides. By using the conventional approach of identifying the antigencontained peptides with highest class II affinity and then examining their capacity to stimulate antigen-specific T cells, we would have had to examine peptides representing two-thirds of the sequence of ␣3(IV)NC1 in order to identify the three nested sets of naturally processed peptides. Better understanding of how antigens are processed is required before current knowledge of class II-peptide interactions can be efficiently used to identify T cell epitopes.